1. Field of the Invention
The invention relates to a method for producing semiconductor wafers of silicon, comprising pulling a single crystal growing on a phase boundary from a melt contained in a crucible and cutting semiconductor wafers from the pulled single crystal, wherein heat is delivered to a center of the phase boundary during pulling, and a radial profile of a ratio V/G from the center to an edge of the phase boundary is controlled, with G being the temperature gradient perpendicular to the phase boundary and V being the pull rate at which the single crystal is pulled from the melt. The invention also relates to defect-free semiconductor wafers of silicon, which can be produced by this method. Semiconductor wafers of silicon in the context of the invention are referred to as defect-free so long as neither OSF defects, A-swirl defects, nor COP defects with a size of more than 30 nm are detectable. A method in the context of this invention is regarded as economically viable when, in particular, it is possible to pull single crystals with a diameter of at least 300 mm from a crucible at a rate which is equal to at least 0.5 mm/min and defect-free semiconductor wafers are produced in a high yield, expressed in terms of the total yield of semiconductor wafers.
2. Background Art
DE 103 39 792 A1 describes a method for producing single crystals of silicon which are optimized with respect to their defect properties. Attention is focused on intrinsic point defects and their agglomerates, as well as the Voronkov model which allows predictions regarding the formation of such defects. In the case of intrinsic point defects, distinction is made between interstitial silicon atoms (interstitials) and vacancies. If point defects enter supersaturation when cooling the single crystal, then silicon interstitials will form agglomerates which can be detected in the form of dislocation loops (A-swirl defects, LPITs) and smaller clusters (B-swirl defects). In the event of supersaturation, vacancies form vacancy agglomerates (voids) which, depending on the detection method, are referred to inter alia as COP defects (crystal originated particles, COPs), FPD (flow pattern defects), LLS (localized light scatterers) or DSOD (direct surface oxide defects). It is necessary to ensure that the semiconductor wafers of silicon have no A-swirl defects in the region relevant for producing electronic components, and are as free as possible of COP defects whose size lies in the range of the structure widths of the components, or are of greater size. Semiconductor wafers which fulfill these requirements are often referred to as defect-free or perfect, even though their crystal lattice generally contains smaller COP defects or B-swirl defects, or contains both defect types.
According to the Voronkov model, the intrinsic point defect type which is incorporated in excess into the crystal lattice when pulling the single crystal depends essentially on the ratio of the pull rate V, at which the single crystal is pulled from the melt, and the temperature gradient G perpendicular to the phase boundary between the growing single crystal and the melt. Often, instead of the temperature gradient perpendicular to the phase boundary, the axial temperature gradient directed perpendicularly to the surface of the melt is also used in model calculations. If the ratio V/G falls below a critical ratio, then an excess of silicon interstitials is created. If the critical ratio is exceeded, then vacancies predominate. If there is an excess of vacancies, the size of the COP defects formed depends essentially on two process parameters, namely the aforementioned ratio V/G and the rate at which the single crystal is cooled in the range of from approximately 1100° C. to 1000° C., the nucleation temperature of voids. The COP defects are therefore commensurately smaller as the ratio V/G lies closer to the critical ratio and the more rapidly the single crystal is cooled in this temperature range. In practice, attempts are therefore made to control the two process parameters when pulling the single crystal, so that the defects created by supersaturation of vacancies remain small enough not to interfere with the production of electronic components. Since the structure widths of the components decrease with each generation, the defect size which can still be tolerated decreases accordingly.
Owing to corrosion of the crucible, usually consisting of quartz, oxygen will enter the melt. The oxygen forms small so-called precipitates in the single crystal (as grown bulk micro defects, BMDs). These are desirable to a certain extent because they can bind (getter) metallic impurities to themselves, and thus can be used in order to move such contaminants away from the region of the surface into the interior (bulk) of the semiconductor wafer.
If the single crystal is pulled under conditions in which the ratio V/G lies only slightly above the critical ratio, then the interaction of vacancies and oxygen atoms also leads to the formation of nuclei, which give rise to OSF defects (oxidation induced stacking faults). The presence of a zone with such nuclei (OSF zone) is usually detected by subjecting a semiconductor wafer, cut from the single crystal, to oxidation in wet oxygen at about 1100° C. for a few hours so that OSF defects are formed. Since this defect type is likewise detrimental to the functional integrity of electronic components, endeavors are made to suppress OSF formation, for example by reducing the concentration of oxygen in the melt so that less oxygen is incorporated into the single crystal than would be necessary in order to form OSF defects. The OSF zone can also be avoided by modifying the ratio V/G, for example by using higher or lower pull rates. The formation of OSF nuclei can moreover be reduced by higher cooling rates (in the temperature range of precipitation around 900° C.). It is furthermore known that in order to avoid OSF defects, it is advantageous for the single crystal to contain a small concentration of hydrogen.
Particular difficulties in controlling the ratio V/G result from the fact that the single crystal usually cools faster at the edge than at the center, so that the ratio V/G decreases from the center toward the edge. Despite corresponding control, this can lead to unacceptably large COP defects being formed at the center and/or A-swirl defects in the edge region. The dependency of G on the radial position r, G(r), must therefore be taken into account, especially when defect-free semiconductor wafers of silicon with sizeable diameters are to be produced economically.
In the aforementioned DE 103 39 792 A1, it is proposed to induce a transport of heat directed from below toward the center of the phase boundary. This is intended to achieve two effects. On the one hand, the increase in the temperature gradient G concomitant with the heat transport is intended to make it possible to increase the pull rate V correspondingly, without defects therefore being generated. On the other hand, it is intended to homogenize i.e. equalize the radial profile of the ratio V/G, so that it varies as little as possible from the center to the edge of the phase boundary and lies as close as possible to the critical ratio. With this strategy, it is feasible to produce defect-free semiconductor wafers with a diameter of 300 mm, in which case the single crystal can be pulled at a rate of 0.36 mm/min.
U.S. Pat. No. 6,869,478 B2 discloses that a phase boundary curved in the direction of the single crystal generates a temperature gradient which is steepest perpendicular to the phase boundary. Taking into account the Voronkov model, according to which point defects diffuse in the direction of the temperature gradient and according to which silicon interstitials diffuse faster than vacancies, it is furthermore disclosed that the radial diffusion of silicon interstitials due to the curvature of the phase boundary increases the concentration of vacancies at the center of the phase boundary. The ratio V/G, at which the concentrations of vacancies and silicon interstitials correspond to each other, will therefore be commensurately less as the phase boundary is curved more strongly toward the single crystal.
The present inventors found that the predictions for defect distributions, even when they take the radial distribution into account, differ commensurately more strongly from the defect distributions found in experiments as the rate at which the single crystal is pulled is faster, and as the diameter of the single crystal is greater.
FIG. 1 shows an extreme example of this observation. A single crystal of silicon with a nominal diameter of 300 mm was pulled at a high pull rate and an inhomogeneous radial profile of V/G was adjusted. In the central region, V/G was adjusted to be so low that the formation of A-swirl defects could be expected in this region according to the predictions of the Voronkov model. In fact, however, COP defects with a diameter of more than 30 nm were found. In the edge region, the ratio V/G was adjusted to be so high that large COP defects should be formed there. In fact, however, A-swirl defects were found.
These results showed that the strategy hitherto followed in the prior art, of adjusting a ratio V/G whose radial profile changes as little as possible and which corresponds as far as possible to the critical ratio, will not be successful when defect-free semiconductor wafers of silicon are to be produced economically.